// use default settings for maxSize, nu, r from state
public static void MakeStep(IGbmState state)
{
var gbmState = state as GbmState;
if (gbmState == null)
{
throw new ArgumentException("state");
}
MakeStep(gbmState, gbmState.maxSize, gbmState.nu, gbmState.r, gbmState.m);
}
// allow dynamic adaptation of maxSize, nu and r (even though this is not used)
public static void MakeStep(IGbmState state, int maxSize, double nu, double r, double m)
{
var gbmState = state as GbmState;
if (gbmState == null)
{
throw new ArgumentException("state");
}
var problemData = gbmState.problemData;
var lossFunction = gbmState.lossFunction;
var yPred = gbmState.pred;
var yPredTest = gbmState.predTest;
var treeBuilder = gbmState.treeBuilder;
var y = gbmState.y;
var activeIdx = gbmState.activeIdx;
var pseudoRes = gbmState.pseudoRes;
var trainingRows = gbmState.trainingRows;
var testRows = gbmState.testRows;
// copy output of gradient function to pre-allocated rim array (pseudo-residual per row and model)
int rimIdx = 0;
foreach (var g in lossFunction.GetLossGradient(y, yPred))
{
pseudoRes[rimIdx++] = g;
}
var tree = treeBuilder.CreateRegressionTreeForGradientBoosting(pseudoRes, yPred, maxSize, activeIdx, lossFunction, r, m);
int i = 0;
foreach (var pred in tree.GetEstimatedValues(problemData.Dataset, trainingRows))
{
yPred[i] = yPred[i] + nu * pred;
i++;
}
// update predictions for validation set
i = 0;
foreach (var pred in tree.GetEstimatedValues(problemData.Dataset, testRows))
{
yPredTest[i] = yPredTest[i] + nu * pred;
i++;
}
gbmState.AddModel(tree, nu);
}
// allow dynamic adaptation of maxSize, nu and r (even though this is not used)
public static void MakeStep(IGbmState state, int maxSize, double nu, double r, double m) {
var gbmState = state as GbmState;
if (gbmState == null) throw new ArgumentException("state");
var problemData = gbmState.problemData;
var lossFunction = gbmState.lossFunction;
var yPred = gbmState.pred;
var yPredTest = gbmState.predTest;
var treeBuilder = gbmState.treeBuilder;
var y = gbmState.y;
var activeIdx = gbmState.activeIdx;
var pseudoRes = gbmState.pseudoRes;
var trainingRows = gbmState.trainingRows;
var testRows = gbmState.testRows;
// copy output of gradient function to pre-allocated rim array (pseudo-residual per row and model)
int rimIdx = 0;
foreach (var g in lossFunction.GetLossGradient(y, yPred)) {
pseudoRes[rimIdx++] = g;
}
var tree = treeBuilder.CreateRegressionTreeForGradientBoosting(pseudoRes, yPred, maxSize, activeIdx, lossFunction, r, m);
int i = 0;
foreach (var pred in tree.GetEstimatedValues(problemData.Dataset, trainingRows)) {
yPred[i] = yPred[i] + nu * pred;
i++;
}
// update predictions for validation set
i = 0;
foreach (var pred in tree.GetEstimatedValues(problemData.Dataset, testRows)) {
yPredTest[i] = yPredTest[i] + nu * pred;
i++;
}
gbmState.AddModel(tree, nu);
}
// use default settings for maxSize, nu, r from state
public static void MakeStep(IGbmState state) {
var gbmState = state as GbmState;
if (gbmState == null) throw new ArgumentException("state");
MakeStep(gbmState, gbmState.maxSize, gbmState.nu, gbmState.r, gbmState.m);
}
